The Yale Law Journal


The Costs of Voting Rule Chevron: A Comment on Gersen and Vermeules Proposal

26 Jan 2007

Professors Gersen and Vermeule argue that we should replace “doctrinal Chevron,” which instructs courts to defer to an agency’s reasonable interpretation of a statute the agency administers, with “voting rule Chevron.” Under voting rule Chevron, judges would not defer to agency views. Instead, voting rule Chevron would induce deference at the aggregate level by requiring a supermajority vote to reverse an agency. Gersen and Vermuele’s argument is novel, provocative, and ingeniously developed. It also has a certain aesthetic appeal: the elusive, imprecise, “soft” Chevron standard is supplanted by a clear, rigorous, “hard” voting rule. Nonetheless, I am not (yet) persuaded of the practical wisdom of the proposal.

A key feature of Gersen and Vermeule’s argument is their claim that a switch from doctrinal Chevron to voting rule Chevron could maintain all the benefits of Chevron deference with respect to case outcomes. The reason, they explain, is that voting rule Chevron can induce approximately the same aggregate amount of deference as doctrinal Chevron (even though, as Gersen and Vermeule acknowledge, the outcomes of some individual cases would change). If Gersen and Vermeule successfully establish that claim, the conversation is no longer about the optimal level of deference, but only about how that optimal level can best be achieved.

Gersen and Vermeule argue that the main advantage of voting rule Chevron (holding the total level of deference constant) is that it avoids the conceptual problems and psychological burdens inherent in doctrinal Chevron. (Gersen and Vermeule claim other benefits as well, which in the interests of space I will not address in this response.) I am skeptical of Gersen and Vermeule’s claim that voting rule Chevron has the advantage of substantially lowering conceptual and psychological costs to judges. It is entirely plausible that doctrinal Chevron, as currently practiced by most judges, places equal or lesser burdens on judges, for the simple reason that deciding whether a legal question is hard, and that an agency’s resolution is within the realm of plausibility, is usually no more difficult—and may often be easier—than deciding whether the agency’s interpretation is, in fact, the single best reading of the statute.

I also take issue with Gersen and Vermeule’s claim that voting rule Chevron and doctrinal Chevron can achieve normatively equivalent outcomes with respect to the level of judicial deference. The nub of my argument, which Gersen and Vermeule acknowledge in passing, is that even if doctrinal Chevron and voting rule Chevron can achieve equivalent aggregate levels of deference, these regimes distribute deference differently across case types. This distributional difference may be normatively significant, even when the aggregate level of deference is held constant. Therefore Gersen and Vermeule cannot, in my view, presume that voting rule Chevron can ever be calibrated so that it produces deference outcomes that are normatively equivalent to those produced by doctrinal Chevron.

I. Conceptual and Psychological Costs

Gersen and Vermeule assert that doctrinal Chevron imposes conceptual and psychological burdens because it requires a judge to make not only a first-order decision about whether the agency’s interpretation or the challenger’s interpretation is more persuasive, but also a second-order judgment about whether the agency’s interpretation is “reasonable” (or “reasonable enough”), even if it is not as good as the challenger’s interpretation. Gersen and Vermeule’s fundamental claim is that deciding which of two options is better is logically prior to, and psychologically easier than, deciding whether the choice itself is hard or easy.

While I am no expert in cognitive psychology, I find this claim dubious. It seems quite common for a decision-maker to find it easier to decide that the answer to a question is not clear—and therefore governed by a background presumption—than to evaluate what the correct decision would be if the presumption did not apply. A batter behind in the count 0-2 knows he should swing at anything close to the strike zone, while a batter ahead in the count knows that he should let a borderline pitch go by. Is it cognitively harder for the batter to evaluate whether a pitch is close enough to the strike zone for him to swing than it is for him to evaluate whether the pitch actually is a strike? Or, consider Gersen and Vermeule’s example, in which a referee is asked to assess whether a scientific study’s conclusions are plausible given the data. The referee may find it easier to decide that the results are plausible than to evaluate whether the results are actually correct. As a final, meta-example, it may be much harder for a reader to decide whether doctrinal Chevron or voting rule Chevron is superior than to conclude that the question is complicated, with reasonable arguments on both sides.

This observation is closely related to the claim that decision-makers often act as “satisficers” rather than optimizers. A satisficer stops when she finds an option that is good enough, rather than trying to figure out which option is the best one. One could conceptualize doctrinal Chevron as an instruction to courts that they should satisfice, and that they should consider the agency’s interpretation first (that is, courts should accept the agency’s view if it passes the satisficing threshold). That conceptualization may not be perfect, but I think it is a workable first-order approximation of judicial review of agency action. Gersen and Vermeule, as I understand their argument, claim that it is conceptually and psychologically more difficult for judges to satisfice than to optimize. That assertion is not self-evidently true and stands in considerable tension with much of the literature on choice under uncertainty.

Gersen and Vermeule do acknowledge the possibility that assessing plausibility may be easier than assessing optimality. Their response to this possibility, however, is puzzling. They assert that they merely seek to compare voting rule Chevron to the version of doctrinal Chevron that currently exists, rather than to other possible versions of doctrinal Chevron. Under current Chevron doctrine, according to Gersen and Vermeule, judges consult “the full panoply of sources to derive statutory meaning at Step One.” Gersen and Vermeule admit the possibility of “a parallel world in which judges quickly glance at the statute to make sure an agency’s interpretation is not implausible,” but they insist that “that is not our world.”

There are two problems here. First, the case for replacing doctrinal Chevron with voting rule Chevron is much less compelling if a straightforward modification to doctrinal Chevron would capture the lion’s share of the benefits associated with voting rule Chevron. If, under current doctrine, judges have to do all the work that they would do even if there were no agency interpretation or no deference principle, but these judges then have to ignore the products of that labor in order to defer to the agency, then this indeed seems like a silly doctrine. But this does not establish the need for replacing a “soft” doctrinal approach with a “hard” voting rule. It only establishes the need to replace a foolish doctrine with a more sensible one.

Second, while I have not done any systematic empirical research on the question, I suspect that most judges inhabit something pretty close to Gersen and Vermeule’s “parallel world.” I think (though I cannot prove) that in a typical Chevron case, the judge looks at a statute, considers the arguments made by the agency and the challenger, and gets a sense of whether the agency’s view is plausible. If the judge concludes the agency’s view is clearly reasonable, she votes to uphold the agency. If the judge concludes that, even on this initial pass, the agency’s view is obviously inconsistent with the statute, she votes to reverse. If the agency’s interpretation is somewhere in the middle—on the outer margins of plausibility, but not obviously lunatic—the judge might devote a bit more effort to thinking about the case. She stops not when she has determined that the agency’s interpretation is the best one, but when she has decided that it falls into the realm of plausibility (that is, when she is satisfied, or “satisficed”). This approach economizes on decision costs. While it does require the judge to make a second-order decision—when to stop thinking—that second-order decision would also arise under voting rule Chevron because the optimizing judge needs to figure out when she has enough information to cast her vote.

Descriptions of Chevron doctrine, including Gersen and Vermeule’s, often make it sound as if judges applying Chevron first determine the single best reading of the statute and only then, as a second step, consider whether the agency’s interpretation strayed too far from that reading. Judges may write their opinions as if this is how they went about reaching their conclusions. But I suspect that it is not the way judges actually decide cases, if for no other reason than that it would be irrationally expensive in light of the costs that Gersen and Vermeule identify. Again, I cannot provide any authoritative evidence that this is how judges decide cases under doctrinal Chevron, so I will leave it to the reader to assess which characterization of the judicial decision-making process is more plausible. If what I have suggested rings true, though, it implies that the shift to voting rule Chevron may not achieve the primary advantage that Gersen and Vermeule ascribe to it, and might even be counterproductive.

Gersen and Vermeule do have another line of response to this concern. They suggest that even if it is easier for an individual judge in an individual case to determine whether an interpretation is reasonable than whether it is best, the aggregate decision costs may still be higher under doctrinal Chevron. Gersen and Vermeule suggest two reasons why this might be so. First, they assert that under current doctrine, courts will eventually have to decide whether an agency’s interpretation is the best possible interpretation; avoiding that question in the first case merely postpones the inevitable. Second, they argue that even if doctrinal Chevron reduces decision costs for judges, it may increase decision costs for other actors, primarily because of the unpredictability of doctrinal Chevron relative to voting rule Chevron.

The first point is, in my view, doctrinally and empirically incorrect. In some cases, it might be clear that the first judicial decision determined that there is only one possible interpretation of the statute. But in other cases, probably most cases, it will be clear that the previous judicial decision held only that the agency’s interpretation was reasonable. In such circumstances, an agency could alter its interpretation from X to Y to X again, and the court applying doctrinal Chevron would never have to decide which interpretation was better as long as both appear reasonable.

Gersen and Vermeule’s second point, about increasing the decision costs of other actors, raises the question whether doctrinal Chevron causes much more legal unpredictability than voting rule Chevron. Space constraints prevent me from addressing this issue in any depth, but my main response to Gersen and Vermeule’s predictability argument is this: just as it may be easier for judges to conclude that an issue is difficult than to determine the right answer, it may also be easier for parties to conclude that an issue is close—and therefore to predict that the agency will win—than for parties to predict whether a supermajority will think the agency’s interpretation is (perhaps only slightly) worse than some alternative. Gersen and Vermeule acknowledge this possibility, but argue that this loss of predictability is not that great. The central question, however, is a comparative one, and neither they nor I have any idea whether the net expected change in predictability caused by a shift from doctrinal Chevron to voting rule Chevron would be positive or negative. We both acknowledge that either is possible, depending on the conditions.

Gersen and Vermeule’s final rejoinder to the suggestion that voting rule Chevron may make life harder for judges is to suggest that if “these matters are unclear . . . the fair-minded conclusion is that the issue of aggregate decision costs probably does not cut strongly in one direction or the other,” and that such concerns are likely trivial in light of the “definite advantages” of voting rule Chevron. But I understood Gersen and Vermeule to be claiming that the “principal advantage” of voting rule Chevron is that it avoids the conceptual and psychological difficulties associated with forcing judges to internalize a deference norm. Unless they are drawing fine distinctions between “conceptual and psychological burdens” (presumably weighty) and judicial decision costs (allegedly trivial), then the argument that the latter ought to be treated as “second decimal” considerations would seem to undermine much (though certainly not all) of the case they advance for switching to voting rule Chevron.

II. The Distribution of Deference

Most normative debates about Chevron doctrine focus on the appropriate level of judicial deference to agency decisions. Gersen and Vermeule cleverly try to sidestep this debate by arguing that the aggregate level of deference achieved by voting rule Chevron can be calibrated so that it is the same as whatever would be achieved by their reader’s favorite version of doctrinal Chevron. I do not think the move is completely successful, however, because the distribution of deference across case types may be normatively significant, even if aggregate deference is held constant.

The distributional consequences of a shift from doctrinal Chevron to voting rule Chevron can be expressed in stylized form as follows. Consider a three-judge panel. Each judge may reach one of three conclusions about an administrative agency’s interpretation of a statute. First, a judge might believe that the agency’s interpretation is correct. Second, the judge might think the agency’s interpretation is obviously wrong. Cases in this second category are “easy reverses.” Finally, the judge might think that though the agency’s interpretation is less persuasive than some alternative, there are reasonable, perhaps powerful, arguments on the other side. Call these cases “close reverses.” (Though it would also be possible to sub-divide the “upholds” into “easy upholds” and “close upholds,” this distinction is not relevant for comparing the two Chevron regimes.)

Under voting rule Chevron, whether the question is easy or close is irrelevant: a judge votes to reverse in both “easy reverse” and “close reverse” cases. The supermajority rule, however, means that the agency wins unless all three judges vote to reverse. Under doctrinal Chevron, the judge votes to uphold the agency in “close reverse” cases, but the ultimate decision is made according to simple majority rule: the agency wins only if at least two judges vote to uphold its interpretation.

The shift from doctrinal Chevron to voting rule Chevron will make a difference in the final outcome in two circumstances. First, when two judges think a case is an easy reverse, but the third would vote to uphold, agency interpretations that would be reversed under doctrinal Chevron are upheld under voting rule Chevron. Second, when all three judges are inclined to reverse, but at least two of the three judges think the case is close, the agency would prevail under doctrinal Chevron but lose under voting rule Chevron. In all other cases, the switch from doctrinal Chevron to voting rule Chevron has no effect on outcomes.

Gersen and Vermeule acknowledge that under voting rule Chevron, agencies will get less deference in “[l]ow-intensity cases” where a supermajority thinks “the case is close, but that the agency’s reading is worse,” but will succeed more often in “high-intensity cases,” where the judges are closely divided and the judges who think the agency is wrong hold this view very strongly. But Gersen and Vermeule conclude that the different pattern of deference associated with voting rule Chevron “seems perfectly consistent with the rationales for Chevron,” whether one looks to the expertise rationale, the democratic accountability rationale, or both. Thus Gersen and Vermeule conclude that this distributional effect is either normatively irrelevant or supportive of voting rule Chevron.

Though Gersen and Vermeule claim that this conclusion is intuitive, I did not find it to be so, and I think one can construct quite plausible arguments for the opposite conclusion. Consider the expertise rationale for Chevron: agencies better understand the practical effects of different interpretations, and Congress thought (or a legal fiction stipulates that Congress thought) agencies should take these effects into account when implementing statutory directives. Expertise does not, however, authorize an agency to disobey congressional commands. A case in which all three judges acknowledge that a case is close—that is, the agency’s interpretation does not clearly contravene the statute—seems like precisely the sort of case in which the expertise rationale supports deference. After all, the expertise rationale presupposes that generalist judges, relying on the traditional judicial tools of statutory interpretation, are likely to get the wrong answer in close cases much of the time, precisely because they lack the expertise that Congress, by hypothesis, thought should be taken into account when determining the statute’s meaning.

In contrast, if a majority of judges think that the agency’s interpretation is clearly inconsistent with the statute—so that reversing is an easy call—then the expertise argument for Chevron deference is much weaker. Congress may have wanted agencies to use their expertise to implement the statute, but agencies’ expertise does not give them a license to ignore the statute. The fact that a majority of judges concluded that the agency is not only wrong but obviously wrong is powerful evidence the agency’s choice cannot be justified by a claim to specialized expertise. This is true even if a minority of judges finds the agency’s interpretation convincing (perhaps just barely).

The claim that the democracy rationale supports the shift in the incidence of deference caused by voting rule Chevron is somewhat more plausible: if judges are closely divided, then perhaps the decision should be made by the politically accountable agency. But this claim is still problematic. A conclusion by a majority of judges that an agency’s legal interpretation is not just wrong, but obviously wrong, suggests that the agency has disobeyed the constraints that Congress meant to impose. The judicial constraint on agency decision-making is undemocratic in one sense, but in another sense it is pro-democratic insofar as it enforces congressional directives. Furthermore, reducing deference in cases in which a judicial supermajority disagrees with the agency but thinks the issue is close seems problematic from a democratic decision-making perspective. If the judges are uncertain whether the statute forbids the agency’s action, then shouldn’t the more democratically accountable agency have a stronger claim to choose?

It is therefore questionable whether it is wise to decrease deference in cases in which a supermajority thinks the agency’s interpretation is wrong but reasonable in order to increase deference in those cases in which a simple majority thinks the agency’s interpretation is obviously wrong. Deference may be most appropriate if a supermajority of judges would agree that the interpretive issue is hard, especially if generalist judges are prone to deciding that hard issue incorrectly. The case for deference may be much weaker if a majority of judges believe that an agency’s interpretation clearly contravenes an unambiguous statutory provision because in such cases the agency’s expertise may be irrelevant from a legal or normative point of view.

For these reasons, I do not understand why Gersen and Vermeule assert so confidently that the “change in the incidence of deference [induced by a switch to voting rule Chevron] is either an improvement or neutral” and that voting rule Chevron performs “equally well” as doctrinal Chevron. Gersen and Vermeule justify this claim by asserting that, under voting rule Chevron, “[r]ather than individual judges setting aside their own views in favor of more expert or more democratic judgments, the voting mechanism enforces respect for agency views at the aggregate level.” But the whole point of this critique is that one cannot look only at aggregate deference. A shift in the incidence of deference from those cases in which the argument for deference is strongest to those cases in which the argument for deference is weakest cannot be counted as an improvement, even if aggregate deference is held constant.

Therefore, in my view, Gersen and Vermeule cannot avoid consideration of the effect of voting rule Chevron on outcomes by asserting that they will select whatever voting rule achieves the same outcomes as existing (or optimal) doctrinal Chevron. If the distribution of deference is taken into account, it will generally not be possible to achieve normatively identical deference outcomes. I hasten to add that Gersen and Vermeule may ultimately be right that voting rule Chevron achieves a normatively superior distribution of deference across cases. I have suggested why I am skeptical of that conclusion, but it is certainly a claim worthy of study and debate.


Gersen and Vermeule’s innovative proposal is most valuable, in my view, because it focuses attention on aspects of Chevron doctrine that are often overlooked. What are the types of cases in which the argument for deference is strongest? How do judges apply the Chevron doctrine in practice? How much effort do they invest in assessing the merits of the various legal arguments? When and how do they decide when they are confident enough in their conclusion to cast a vote? These are vital questions, not just for assessing proposals like Gersen and Vermeule’s, but for the study of administrative law more generally.

Whatever the merits of Gersen and Vermeule’s Chevron voting rule idea as a practical proposal—a question on which we disagree (at least for now)—the proposal is fascinating as an intellectual provocation. The more general question the proposal raises, which I hope the authors explore in future work, is the purpose of legal doctrine more generally. If Gersen and Vermeule are right that many of the results that the legal system generates through doctrinal solutions can be achieved through voting rules or other “hard” institutional solutions, then what is the purpose of legal doctrine, and what explains its stubborn persistence as a means of allocating decision-making power? I suspect that more inquiry along these lines may uncover advantages to doctrinal strategies of power control and allocation that have not yet been fully appreciated.

Matthew C. Stephenson is Assistant Professor at Harvard Law School. He wishes to thank Jacob Gersen, Daryl Levinson, John Manning, and Adrian Vermeule for helpful comments on earlier drafts of this reply.

Preferred Citation: Matthew C. Stephenson, The Costs of Voting Rule Chevron: A Comment on Gersen and Vermeule’s Proposal, 116 Yale L.J. Pocket Part 238 (2007),

Related Content:

  • Article: Jacob E. Gersen & Adrian Vermeule, Chevron as a Voting Rule, 116 Yale L.J. 676 (2007).
  • Response: Richard J. Pierce, Jr., Chevron Should Not Be Converted into a Voting Rule: A Response to Gersen and Vermeule’s Proposal, 116 Yale L.J. Pocket Part 248 (2007),
  • Response: Matthew C. Stephenson, The Costs of Voting Rule Chevron: A Comment on Gersen and Vermeule’s Proposal, 116 Yale L.J. Pocket Part 238 (2007),